REVIEWING.md

How to review or attack TECT in 30 minutes

TECT is organized so that a reviewer can start from current state, locate the smallest authority needed for a question, and reproduce or falsify a claim without loading the historical corpus.

1. Start with bounded current views

Open these compact generated pages first:

• management/INDEX.md: live tasks, cited gates, and reader routes.
• CLAIMS.md: current claim tiers, scopes, hypotheses, and falsifiers.
• theory/proof-evidence/INDEX.md: counts and targeted proof-evidence search commands.
• claims/GATES-INDEX.md: live card-to-gate and card-to-hypothesis references, including source mismatches.

The large ledgers remain compatibility and audit authorities. Open them only at the linked anchor or search them by ID.

rg -n "<claim-result-gate-or-exploration-id>" theory/proof-evidence-map.md
python verification/scripts/exploration.py search --claim <ID>
python verification/scripts/exploration.py search --verdict failed

An exploration verdict is non-tier-bearing: advanced is not a proof, and failed is not a global no-go unless a formal negative authority says so.

2. Validate the repository

For a quick claim-ledger check:

python verification/scripts/lint_claims.py

For the publication gate used before every repository commit:

python verification/scripts/release_check.py

Exit zero means the generated views match their authorities and the registered verification gates pass. It does not promote the mathematical or physical tier of any claim.

3. Read one claim at its registered scope

Open claims/<ID>/claim.md and claims/<ID>/status.json. Check, in order:

1. the exact statement and pinned scope;
2. the maturity tier and lifecycle;
3. named hypotheses and open gates;
4. the falsifier;
5. the reproduction command and expected output;
6. the proof note, run artifact, and adversarial review linked by the card.

T5 means closed only within its pinned scope. T6 means a theorem modulo the listed hypotheses. Neither label establishes a broader physical interpretation.

4. Find the current weak points

Do not rely on a hand-written list of "current weakest joints" in a long-lived document. It becomes stale as gates close or priorities move. Use:

• management/INDEX.md for live work;
• claims/GATES-INDEX.md for card-cited gates and registry/card mismatch warnings;
• negative-results/INDEX.md for recent failed routes and audit findings;
• todo.py list --status in_progress, then --status next and --status blocked, for the exact task queue.

The full authorities remain claims/GATES.md, negative-results/registry.md, and todo/todo.json.

5. Falsify or reproduce a claim

Use the card's own falsifier and reproduction package. A counterexample must match the registered domain, conventions, regulator, and parameter ledger. If it breaks a named hypothesis rather than the theorem under that hypothesis, report that distinction explicitly.

Confirmed failures are recorded in the negative-result registry with evidence and credit. Reproduction environment requirements are defined in governance/verification-standard.md.

6. Reporting

Report the claim ID, exact command or construction, environment, observed output, and the narrowest conclusion supported by it. Review rounds live under reviews/; confirmed corrections are linked through reviews/errata/ and the append-only changelog.

Legacy evidence pointers and prose that exceeds a registered card are not authoritative. The current claim card and its linked proof authority win.